About Me

I am a Senior ML Engineer working on production GenAI systems, with a focus on evaluation, reliability, and LLMOps. At IABG, I lead and build systems such as chat assistants, RAG-based document search, meeting note-taking, and code-review automation. Much of my work is about making these systems measurable: defining evaluation criteria, monitoring product and model behavior, and deciding when a system is ready to move from test to broader rollout.

Before that, I completed a PhD at the Technical University of Munich on AI-based failure management for large-scale cloud systems. That research focused on failure prediction, root-cause analysis, and operational risk classification using logs, metrics, and other operational data. It still shapes how I think about ML systems: define the expected behavior, measure deviations, monitor the right signals, and use that evidence to improve reliability, safety, and deployment decisions.

Outside my day-to-day work, I like projects where ML meets a concrete system or interaction: reinforcement learning and games, IoT and home automation, computer vision for physical-world workflows, and NLP or language analysis. I tend to enjoy the full loop: modeling the problem, building the software around it, instrumenting it, and seeing where it breaks.

For contact, use email or LinkedIn.

Core Capabilities

Production-ready GenAI & LLMOps

LLM eval harnesses LLM APIs Guardrails RAG Hybrid retrieval Embedding models LoRA / PEFT LangChain HF Transformers vLLM

Applied Machine Learning

PyTorch sklearn OpenCV NLTK PyG / GNNs RL Gym Time-series analysis Tabular foundation models Pattern mining

MLOps, Platforms & Observability

MLflow DVC W&B FastAPI Docker Kubernetes / Helm ArgoCD GitLab CI/CD GitHub Actions Prometheus Grafana Elasticsearch Kibana

Software Engineering

Python SQL Bash C++ Poetry / uv Ruff ty Pylint pytest

Security, Privacy & Governance

Confidential computing Differential privacy ML privacy attacks Threat modeling AI governance

Experience

Senior ML Engineer, IABG

October 2023 — present

  • Build and maintain an LLM evaluation platform/library for compliance review in regulated settings and continuous delivery of GenAI products, with a multi-dimensional taxonomy covering correctness, grounding and attribution, retrieval quality, safety and toxicity, privacy leakage, security, latency, and robustness.
  • Build and ship production GenAI services including chat assistants, RAG-based document search, and code-review automation, with retrieval pipelines, prompt and tool orchestration, structured outputs, input/output guardrails, vLLM serving on Kubernetes, versioned CI/CD, and canary rollouts.
  • Define KPI/SLA-driven monitoring for GenAI products with Grafana/Prometheus, and use post-incident reviews to close operational gaps.
  • Project lead of SPARTA, an academic collaboration project on social-media monitoring and analysis, with work on political stance detection and malicious-account detection using vision-language embeddings and graph neural networks.
  • Tech lead & core developer for RESISTANT, a zero-trust avionics research project, designing attack-graph modeling via partially-observable Markov decision processes, particle-filter-based risk assessment, and TabPFN-based tabular modeling for adaptive, security-critical aircraft systems.
  • Engineer privacy and safety measures for sensitive AI deployments, including differential privacy, homomorphic encryption, trusted execution environments, and hardening against membership-inference, extraction, and inversion attacks.
  • Mentor colleagues on research directions, clean-code practices, evaluation design, and architectural decisions for reliable and secure GenAI systems.

    • PhD Candidate, Huawei Munich Research Center + Technical University of Munich

    January 2020 — September 2023

  • Researched AI-based failure management for large-scale cloud systems, covering online failure prediction, root-cause analysis, and operational risk classification.
  • Published the first-ever survey on AIOps methods for failure management (ACM TIST), organizing the field by data requirements, intervention windows, and target problems.
  • Built LogRule, a structured-log root-cause analysis method published in IEEE TNSM, achieving 37x faster runtime and improved explanation accuracy.
  • Developed reliability and failure-prediction models for cloud infrastructure, including work on optical transceiver reliability recognized as an IEEE/ACM CCGRID Best Paper nominee.
  • Developed an LLM-based command-line risk classifier for operational security workflows; lead inventor on the related filed patent.
  • Mentored students through seminars, master theses, and guided research projects.

    • Education

    TUM
    PhD, Computer Science

    Technical University of Munich · 2024

    AI-based proactive failure management in large-scale cloud environments.

    TUM
    MSc, Computer Science

    Technical University of Munich · 2019

    Machine learning, deep learning, computer vision, and NLP.

    PoliTo
    BSc, Computer Engineering

    Politecnico di Torino · 2017

    Computer engineering foundations across software, systems, and applied ML.